I'm interested in how technology expands what people can understand, create, and accomplish. I value independent thinking, thoughtful design, and work that grows more useful over time.
Selected Projects · Developer Tools · Skill Set
Six projects spanning AI products, data workflows, cloud optimization, and developer tools.
A pet companion prototype bringing character-based AI chat, pet browsing, bookings, and merchant workflows into one product. A separate scoring service tracks safety signals across rolling time windows; the checkout experience is simulated.
Inside: React, Express, MongoDB, and Hugging Face integrations. Persistent chat history and time-weighted scoring connect the conversation to a broader product journey.
Source · Scoring implementation
See the product interface
A deployed launch verification platform I designed and built for AI-built applications. Truna turns product requirements into controlled browser checks, evidence-backed findings, and repair handoffs for coding agents.
Inside: TypeScript, Next.js, Express, Prisma, and Playwright. Reviewed verification plans connect browser execution with screenshots, network signals, risk analysis, and project-level launch reviews.
Open Truna · Product and Architecture
See the product interface
A discovery hub that turns builder news into a daily reading list. It collects topics from sources including Hacker News, GitHub Trending, Product Hunt, Reddit, and RSS, then ranks recent items using community activity and recency.
Inside: React, FastAPI, and SQLite. Source caps and deduplication keep the daily list varied; optional AI enrichment has template fallbacks, with a review queue for moderation.
Source · Ranking implementation
A ticket classification toolkit that runs a fine-tuned Laya model locally or calls Jev through an API. Use the Python SDK, connect a helpdesk to the REST service, or work from a shared inbox.
Inside: Python, PyTorch, and Docker. Model adaptation and held-out comparison sit alongside workspace access controls, idempotent imports, and Zendesk sync. Source and Python package downloads are available.
Source · Model Card · Downloads
See the ticket inbox with synthetic data
A Windows companion that keeps Codex usage limits and token activity within sight. Its floating HUD expands into daily breakdowns, history, reset reminders, and API-equivalent cost estimates.
Inside: Tauri, Rust, React, and TypeScript. Session history is parsed locally; the model pricing catalog updates independently and has an offline fallback. Windows installers are available.
See the desktop app with demo data
An AWS Lambda cost optimizer that recommends configurations against latency targets. Analyze benchmark results, CloudWatch metrics, and cold-start signals, or start with a local simulation that needs no AWS credentials.
Inside: Python, boto3, and CloudWatch. Cost models, Pareto analysis, and SLO risk scoring turn benchmark data into explainable configuration recommendations. Currently in beta.
Source · Quick Start · Recommendation Logic
Self-hosted memory for AI coding assistants. Save project decisions and conventions, retrieve them from another tool through MCP, or export a handoff briefing.
Inside: Python, PostgreSQL, and pgvector, with scoped client access, an owner review inbox, and backup tools. Currently an alpha project.
| Area | Technologies |
|---|---|
| Languages | Python, TypeScript / JavaScript, Rust, Go, SQL |
| Web, Mobile & Desktop | React, Next.js, React Native / Expo, Tailwind CSS, Tauri |
| Backend & Messaging | FastAPI, Node.js / Express, Socket.IO, Redis, Celery |
| Databases & ORM | PostgreSQL, MongoDB, SQLite, SQLAlchemy, Prisma, Mongoose |
| AI & Retrieval | MCP, RAG, pgvector, OpenRouter, PyTorch, Hugging Face Transformers |
| Cloud & Delivery | AWS Lambda, IAM, CloudWatch, boto3, Docker / Compose, GitHub Actions, Vercel |
| Auth & Validation | OAuth 2.0, JWT, Clerk, Pydantic, Zod |
| Testing & Quality | Playwright, pytest, Vitest, Jest, Ruff, mypy, CodeQL |